Does the number of nodes/elements affect the results?

[ismailqeshta]

Hi,

I am facing a problem in my pushover analysis. When I increased the number of nodes within the column from 12 to 22, I got a continuous linear curve, although I get exactly the same periods for both models.

I am modelling the columns using "element nonlinearBeamColumn".

Does the problem emerge from the number of elements? or it is from the number of nodes?

If it is from the elements, is there any way to connect one element through more than two nodes?


Regards.

[EricsonEncinaZ]

Could you expand on your question. It is quite hard to understand your problem and the source of it.

[ismailqeshta]

Hi EricsonEncinaZ,

Many thanks for your reply.

Yes. For example, models A and B show different pushover results, although they are exactly the same in terms of geometry, materials, fixity, boundary conditions and applied loads. Model A basically shows a continuous linear behavior. The models are only different in terms of the number of nodes used to define each one. As shown below, Model A has 22 nodes, while model B has only 4 nodes.

Column model A:

node 401 0 8000 100
node 402 0 8024.786325 100
node 403 0 8074.774775 100
node 404 0 8124.761905 100
node 405 0 8174.747475 100
node 406 0 8224.731183 100
node 407 0 8274.712644 100
node 408 0 8324.691358 100
node 409 0 8374.666667 100
node 410 0 8424.637681 100
node 411 0 8474.603175 100
node 412 0 8524.561404 100
node 413 0 8574.509804 100
node 414 0 8624.444444 100
node 415 0 8674.358974 100
node 416 0 8724.242424 100
node 417 0 8774.074074 100
node 418 0 8823.809524 100
node 419 0 8873.333333 100
node 420 0 8922.222222 100
node 421 0 8966.666667 100
node 422 0 10000 100

Column model B:
node 401 0 8000 100
node 402 0 8024.786325 100
node 403 0 8074.774775 100
node 404 0 10000 100

[EricsonEncinaZ]

If you are doing an elastic analysis it should not matter the number of nodes and elements, but if your analysis contains non-linearities (and depending on the type of elements and loads) meshing is quite important. If this is your case, this is the reason why you have different results. A meshing sensitivity analysis should be performed to determine the best meshing scheme that provides accurate results and it is not too computationally expensive.

[ismailqeshta]

Hi EricsonEncinaZ,

Many thanks for your valuable answer. Yes, I am running non-linear analysis.

Apart from the mesh sensitivity, I actually need to create many nodes within the column to apply a uniformly distributed load as discrete point loads. Is there any way to create nodes for the point loads, without re-meshing the model?

Regards.

[EricsonEncinaZ]

eleLoad command (http://opensees.berkeley.edu/wiki/index ... ad_Command) could be an option.
Why do you need to create as many nodes?
Pay special attention to the elements that you are using in your models. Basically, force-based elements assume linear moment distribution along the element and displacement-based ones assume linear curvature along the element (this is a simplified version of the force and displacement fields assumed for either element types). You have to provide a mesh scheme appropriate to capture the NL response according to each element's capabilities.

[ismailqeshta]

Hi EricsonEncinaZ,

Many thanks for your reply.

I am using the force-based elements.

The reason I need to provide as many nodes is that I want to control the magnitude of each single point load. If I use the eleLoad command, I will not be able to do so, as it only provides the uniformly distributed load over the element length.

[selimgunay]

Why do you need to apply many point loads on a column? Is it a distributed system and you are doing pushover analysis? or is there another reason?

[ismailqeshta]

Hi selimgunay,

Yes. There is actually another reason. I want to apply point loads at different height within the same model.

Is there any way to do it without affecting the mesh of the column?

[selimgunay]

There is no other way if you need to apply point loads. If it is okay for you to apply distributed loads, you can use the eleload command and would not need to divide the element into several small elements.

[ismailqeshta]

Hi selimgunay,

Thanks for the suggestion.

Did you mean that I can select the location of the distributed load within the column height? For example, if I have a column of total height of 10 m, can I specify the distributed load to be applied at height of 5 distributed within 1 m?

[venkateshyadav]

EricsonEncinaZ wrote:
> If you are doing an elastic analysis it should not matter the number of
> nodes and elements, but if your analysis contains non-linearities (and
> depending on the type of elements and loads) meshing is quite important. If
> this is your case, this is the reason why you have different results. A
> meshing sensitivity analysis should be performed to determine the best
> meshing scheme that provides accurate results and it is not too
> computationally expensive.

Hi EricsonEncinaZ,
As you said for non linear analysis, is meshing important? You mean meshing means a single beam element is divided into more number of nodes and more number of elements. But how to connect those nodes? Is this a problem for non convergence while using forceBeamColumn element for RC frame pushover analysis?
Thanks in Advance

[EricsonEncinaZ]

Hi,
In order to define a beam-column element you need 2 nodes, when one node is shared between 2 or more elements then those elements are connected through the common node, so continuity of the element is automatically achieved.

Lets assume that this -- represents an element and this o represents a node:
o--o--o: 2 elements defined using 3 nodes, since the node in the middle is shared the 2 elements represent a continuous element
o--o o--o : 2 elements defined using 4 nodes, there is no common node, so the elements are independent between each other. (Unless you use equalDOF between the central nodes to pair their DOFs)

Now, about your questions on whether that affects convergence, it may affect it, everything depends on your problem and how you set up your model. You have to know the capabilities of the materials, sections, elements and modelling strategy/technique you are using.

[selimgunay]

You don't need to divide the force based beamcolumn elements into smaller elements, for the displacement based elements you do need that. Please see

http://opensees.berkeley.edu/wiki/index ... ed_Element

[venkateshyadav]

Thanks for your suggestion EricsonEncinaZ and selimgunay.
So, if I model my element into number of nodes and use the dispBeamColumn element, will it be almost accurate ? when compared with the forceBeamColumn single element?
I have read that file. Is it true for all the cases?
Is there any procedure so that I cannot come again with convergence issues?
Thanks